Comparison of Estimated SARS-CoV-2 Seroprevalence through Commercial Laboratory Residual Sera Testing and a Community Survey

Abstract We compared severe acute respiratory syndrome–related coronavirus-2 seroprevalence estimated from commercial laboratory residual sera and a community household survey in metropolitan Atlanta during April-May 2020 and found these two estimates to be similar (4.94% versus 3.18%). Compared with more representative surveys, commercial sera can provide an approximate measure of seroprevalence.


Introduction
In the setting of widespread community transmission of severe acute respiratory syndrome-related coronavirus-2 (SARS-CoV-2), serological surveys are important to inform the public health response.
They can help to estimate infections at the population-level and inform transmission dynamics [1]. By detecting evidence of past infection, serological surveys can identify more infections than reported cases alone. This is particularly useful where testing is limited and for persons with mild or asymptomatic infections who may not be tested.
Many SARS-CoV-2 serological surveys are being conducted, but vary by participant and sample selection methods, assay type, period of study, and regional differences in the burden of disease. Carefully designed studies to select participants representative of the underlying population may provide the most accurate seroprevalence estimates but require considerable investment of time and resources [2,3]. Conversely, convenience sampling can facilitate large, efficient studies and be used to follow seroprevalence trends over time, though the population from which specimens are obtained may not be representative of the general population [4][5][6][7][8].
To better understand how sampling strategies impact estimates of coronavirus disease 2019 (COVID-19) burden and to assess the ability of commercial laboratory sampling to accurately estimate SARS-CoV-2 population seroprevalence, we compared the seroprevalence estimate from a convenience sample of commercial laboratory residual sera ("commercial sampling") with the estimate from a community survey designed to randomly select a representative sample of metropolitan Atlanta residents ("community survey") [3]. For this comparison, we used the same serologic assay and matched overlapping geographical areas and time periods. With the commercial sampling, we also estimated seroprevalence at multiple timepoints from April-July 2020 in the

Survey Populations
In commercial sampling, persons of all ages throughout the Atlanta MSA, which includes DeKalb and Fulton Counties, had blood testing through a commercial laboratory from April 27-May 1, 2020 for routine screening or clinical management. Additional testing was also performed in the Atlanta MSA from May 23-30, June 15-17, July 6-10, and July 27-28. A convenience sample of deidentified specimens was selected by the laboratory for further SARS-CoV-2 testing. We targeted at least 300 specimens per age group [5]. The community survey was conducted in Dekalb and Fulton Counties from April 28-May 3, 2020. Three hundred ninety-four households and 696 participants of all ages were enrolled using a stratified two-stage cluster sampling design as previously described [3].
This activity was reviewed by CDC and was conducted consistent with applicable federal law and CDC policy. 1

SARS-CoV-2 Antibody Testing
For both investigations, SARS-CoV-2 antibody testing was performed by the Centers for Disease Control and Prevention (CDC) using a CDC-developed and validated enzyme-linked immunosorbent assay (ELISA) with pan-immunoglobulin against the anti-spike protein of SARS-CoV-2. Through methods previously described, specificity was 99.3% (95% CI: 98.3-99.9%) and sensitivity was 96.0% (95% CI: 90.0-99.9%) [5,9]. To compare seroprevalence estimates using the same assay, serum samples in the community survey originally tested with a different assay [3] were re-tested using the CDC ELISA. M a n u s c r i p t 5

Analysis
We compared age, sex, and available race/ethnicity data in the two populations with demographics of the corresponding catchment areas using U.S. Census data [10]. For commercial sampling, we identified residents of the Atlanta MSA by ZIP codes contained in the 29-county catchment area using population density mapping. Seroprevalence results were age-and sex-standardized to the underlying population and adjusted for assay characteristics using previously described methods [5]. For this analysis, we report the final adjusted seroprevalence estimate as the mean of the bootstrap distribution.
We repeated analysis of the community survey using test results from the CDC assay by methods previously described [3]. For this analysis, we also adjusted the seroprevalence estimates to account for the assay characteristics [9], uncertainty in the test positivity from the community survey, and uncertainty of the sensitivity and specificity of the assay [11].
We examined whether there was a significant difference in seroprevalence estimates between commercial sampling and the community survey conducted in similar geographic areas during the same time period. We calculated the effective sample sizes using a binomial distribution with the corresponding mean and variance from each study and used a two-sample z-test to compare the estimates.
Finally, we calculated seroprevalence estimates for commercial specimens collected in the Atlanta MSA during three additional time periods from May through July and compared changes to

Discussion
The validity of large-scale SARS-CoV-2 seroprevalence surveys utilizing clinical blood samples relative to survey methods traditionally more representative of the population has not been evaluated.
We compared SARS-CoV-2 seroprevalence estimates from commercial laboratory residual sera testing with a community survey in metropolitan Atlanta and found estimates to be similar. The general trend in seroprevalence estimated from commercial sampling was also consistent with the pattern of reported cases in the Atlanta MSA.
Seroprevalence surveys involving commercial residual sera testing can estimate the total number of SARS-CoV-2 infections across many different geographic areas. Furthermore, the ability to efficiently conduct repeated sampling over time has allowed CDC to track jurisdiction-level spread of SARS-CoV-2 across the U.S. [8]. Despite these important advantages, such convenience sampling may be subject to bias, as persons who have their blood drawn for routine clinical purposes may differ systematically from the general population with regard to demographics, geographic distribution, A c c e p t e d M a n u s c r i p t 7 underlying health status, exposure risk, care-seeking behavior, and healthcare access [5]. Even with these biases, we found that commercial sampling provided an efficient means to estimate community SARS-CoV-2 seroprevalence relative to the community survey conducted during the same period in approximately overlapping geographic areas.
Many factors contributed to overall differences in the seroprevalence estimates, but we could not quantify their relative contribution or direction of bias in all cases. Local variation in SARS-CoV-2 outbreaks can impact seroprevalence estimates, particularly when measured over small geographic areas. While we observed early COVID-19 incidence to be higher in DeKalb and Fulton Counties compared with the entire Atlanta MSA, we were not able to compare seroprevalence in the two counties alone as the commercial sampling was not powered to examine county-level seroprevalence.
This study was also not powered to detect less than a 5% difference between commercial sampling and the community survey. In addition, risk of SARS-CoV-2 infection and seroprevalence have been shown to vary across age and race/ethnicity groups [2,3,5,8,13] and can therefore impact seroprevalence estimates. Though individual-level race information was not available for the commercial sampling, the geographic distribution of the residual specimens within the Atlanta MSA (Supplementary Figure S2) suggests that commercial sampling resulted in underrepresentation of racial and ethnic minorities. Minority underrepresentation was also observed in the community survey [3], but without individual-level race information, we could not directly compare differences between the two surveys. Differences in the geographic areas as well as the populations sampled could account for the difference in seroprevalence estimates in the two studies.
In conclusion, we found estimation of SARS-CoV-2 seroprevalence from commercial laboratory residual sera testing to be a reasonable alternative approach to a community survey in metropolitan Atlanta. Given the limitations of commercial testing, more resource-intensive approaches may still be required to ascertain more accurate seroprevalence estimates in special populations, including minority groups, children, and healthcare workers [14]. Large-scale geographic seroprevalence surveys offer important advantages to allow tracking of SARS-CoV-2 spread through the U.S. M a n u s c r i p t 11 Table 1.